001 - Redesigns, Conferences, and Rogue Agents
Hello, and welcome to the Thinking Peter podcast where we cover the news of the community and learn from
each other.
My name is Peter Ulrich.
And I am Gus Workman.
Are you Gus Workman, are you Peter Ulrich as well? I might be Peter Ulrich.
Yeah.
This is just me with a different voice and a different haircut and a different face.
Otherwise, it's just me and and a clone.
No.
Anyways, this was a little joke.
Yeah.
My name is Peter Ulrich, and my good friend, Gus Workman, he and I, we decided to just do this little
podcast thing and see whether we can contribute a little bit to the Elixir community by giving you more
or less weekly, biweekly news, you know, what's big and what's important right now in the industry and
our language.
And, yeah, it's definitely not thinking Elixir two point o, you know.
We very much appreciated the the original thinking Elixir, and we couldn't even try to step into their
very big shoe sizes of shoe.
What do you call it? Like, the shoes.
Right? Couldn't fill those shoes.
Not not if you tried very hard.
So this They're definitely gonna be missed.
They will be missed, and I hope they do come back, and then we can have a very friendly competition.
I hope you're nice.
Now this podcast is called macro mayhem.
Gus and I came up with this last Friday, and today is Monday, the twenty seventh of July.
So here we are.
I think we're living up to the name in terms of our whirlwind speed.
Exactly.
Yeah.
We were just iterating, and and here we are.
So macro mayhem, what we're gonna do what we're gonna try to do is to have a biweekly release schedule.
So every two weeks, you will have a new episode in your podcasting app.
We will try to structure the podcast in a way that we first give you a couple of highlights or new item
news items that might interest you.
We're gonna discuss them very briefly, not, you know, super big, just like maybe a comment or two so
that you get interest you.
There's links.
It yeah.
We will put all the links in the show notes.
So but the idea here is that we give you kinda like a, you know, fifteen minute, twenty minute overview
of what's currently happening, what has happened in the last two weeks.
And then after we have done the news, we're gonna do a bit more broader discussion section where we take,
you know, bigger news items that happened, and we discuss them, give our opinions on that that kind of
stuff.
So for you as a listener, you know, if you're only interested in the news, you can just turn off the
podcasting app after fifteen, twenty minutes, and don't bother about, you know, two old gentlemen rambling
about what, you know, their opinions on on current affairs.
But, yeah, that that's basically it.
One last item.
If you if you wanna have the your work or your blog post, whatever.
If you wanna have it featured on the podcast on this podcast, you know, I can only work on this one.
But if you wanna have it featured, just tag us, me or Gus.
So I'm Peter Ulrich.
Gus is at Gus Workman on Blue Sky.
G workman.
G workman.
Yeah.
Yeah.
Top g workman on Blue Sky.
You can also find us on LinkedIn if you don't have blue sky.
Otherwise, I'm on the Elixir forum.
So if you have a post there, just tag me and then, you know, we can't promise you that Both are? Yeah.
We both are.
So, yeah, tag us both ideally.
And on the on the Elixir forum, then, you know, we can't promise you that we will feature it because
we need to prioritize for time.
But, definitely, it it increases your chances of, you know, being mentioned on this podcast.
Alright? Gus, who are you, actually? Who who are you? Why are you here? Who am I? That is a good question.
I feel like a lot of people know you, Peter.
But maybe maybe I'm coming out of the shadows here.
Yeah.
I'm Gus Workman.
I am a I've been involved in Elixir for like, oh, maybe eight years now.
I think I started in twenty eighteen.
Ah, you started before me.
But most of my work has been on the kind of nerves side of things.
Right.
So I'm not huge in phoenix circles.
I I I do phoenix a lot, but usually in the context of nerves.
So I'm a host for the Nervous Meetup.
It's a remote online meetup.
I've been around at Gothmeyer last year helping out with some workshops this year, done some conference
speaking at ElixirConfU.
Again, mostly in the hardware nerve space.
I actually have a hardware background.
I actually I think I think what most people know you for is the the name badge from Goatmeier.
Right? Yeah.
So tell us about that.
Like, will there be name badge two point o this year? Yeah.
So if you're at Gottmire, I'm not gonna spoil anything, but we're working on some hardware ideas, exploring
stuff.
And when it makes sense, there may or may not be some hardware at the event.
There's definitely hardware at the workshops beforehand.
Right.
So there's gonna be a lot of fun stuff.
But I don't know.
Gottmaya is just gonna be a fun time for nerves.
We we will talk about Gottmaya in a bit.
And if you are not coming, you will miss out probably the best conference in years.
And unfortunately, also the last one, the last Goldmire for at least this next year.
Like, next year, there won't be a Goldmire.
So if you wanna go go this year, just, you know, before you you miss the good stuff.
But I guess you said mentioned earlier you mentioned earlier, like, why are you so focused on hardware?
Because everyone's just doing software and web apps these days, isn't it? Well, it's just it's what I
love.
I my education background, I I have a degree in computer engineering is what my university called it.
Right.
And that really was the intersection of hardware and software.
I took classes in hardware design as well as like CS classes.
So I have that hardware background and I've always just loved working with my hands.
Mhmm.
I love the feeling of like I built something, can touch it and feel it and then also understanding how
it works under the hood.
Right.
So that's kinda where the software part of things come in.
And and where did you maybe last before we move into the news section, like, where where did you work
before you switched over to Nervous? Like Before I switched to Nerfs? Like, how did you get started on
hardware? I was working at Samsung prior to moving to Europe and Right.
Starting my own hardware, software, nerves consultancy.
So I had a lot of experience in r and d development stuff there, that was a lot of fun.
So that's why I decided to continue.
Yeah.
Awesome.
Well, you know, you're the the the hardware guy.
I'm the the software guy.
So hopefully, we can make a whole pie.
Let's move into the news.
Right? That's what people are here for.
So first up, well, no, I shouldn't say first up.
That sorry.
That's thinking Elixir.
Like, I've been listening to that show for six years, so everything's just, you know, ingrained in my
brain.
You know, these these things.
With you.
K.
Let's find something else.
The first item on our news agenda here is the release of Hex two point five point o.
It's already two, three weeks ago, but, you know, in in software terms, that's ages ago.
But I think it was a very big and important release.
And the biggest thing that you might have noticed is that the mixed steps get command got an upgrade,
and it now lists vulnerabilities that have been reported to HEX.
So if there's a vulnerability on HEX and you run mixed steps get, you will get warnings about these vulnerabilities.
They are not blocking you, I believe.
Well, they're still fetching the the dependencies.
Right? But you see a warning.
That that is an important one.
The other important one is that there is also an update to mix hex audit, and that is more or less a
good replacement for the Mix audit, the Mix audits package, which is kinda like a third party library.
And Mix audit, they they always pulled in vulnerabilities as well, and they would crash your CI or they'd
fail your CI if there was a dependency with the vulnerability.
Right? But that kinda got moved into HEX audit now.
And HEX audit uses the vulnerabilities from the HEX package manager.
Whereas Mix Audit, they used it from, like, OVS.
There are some, you know, big platforms where all the vulnerabilities get reported.
And GitHub advisories, but they might always be a little bit behind than what Hex has.
So, yeah, just for you as a heads up, if you upgrade your your local mixed steps get, like, your mix
hex locally, mixed steps get will show warnings.
And if you have mixed audit, you can replace it with mixed hex audit.
And yeah.
Because I'm also busy with, like, vulnerability stuff, I know that there will be more work done on mixed
hex audit in the future.
So, you know, switch before rather now than later.
Yeah.
I think I also saw that they added the timeouts to or, the cool down period to Hex.
I know a lot of people were asking about that, and there was some discussion as to whether or not we
need that in the Elixir ecosystem, but I think that was something that if people are interested in, they
should look into that as well.
Right? Yes.
Did they release it now already? I I know there's a discussion about when you release a new version that
is gonna be held back for twenty four hours before people can actually update or download it, you know,
like, at least automatically.
I'm not sure whether they landed that one already, but, you know, we'll get back to that next week or
in two Cool.
Alright.
Well, Gus, what what's next on the items? Next was the elixir lang dot org redesign.
This was a I I think the updates look great but if you didn't see it, go to elixir lang dot org and you
can see the new website for the Elixir language as a whole.
The Elixir team updated it.
It looks really nice, an updated modern design, a new logo, um, and in general, I think it just modernizes
the language and makes it a little bit more approachable or the appearance of being the freshest, most
modern up to date stuff for newcomers who haven't seen it before.
Yeah.
What what I what I like about it most is that they also added new use cases of Elixir.
So previously, it was mostly focused on the language and now they use, well, Phoenix web development.
They have embedded systems, so nurse.
They have machine learning where they highlight numerical Elixir and X, and then also LiveBook and the
Bundleby and Axon combination.
They do have a section for data and media, so you use Broadway and Membrane.
And then they have one for IoT and distributed systems where they basically mention the EMQX.
Maybe I never heard about that one, to be honest.
But it's, you know, they think thanks to the Erling VM, Alexei also excels at distributed systems and
everything Erlang is renowned for.
So RabbitMQ, React.
Yeah.
So I think it's mostly distributed systems that they highlight.
Yeah.
And it's great that they're highlighting the entire ecosystem because Elixir doesn't exist in an isolated
environment.
It's everything that all the mature libraries and ecosystem that surrounds it is what really makes our
our community really great.
Yep.
True.
And especially for for a new user who comes in, they might not understand that they, you know, can build
websites, but they can also program hardware, and they can also do data science, and they can also do,
you know, IoT, that kind stuff.
So it's it's a great Right.
Update.
And then the next website redesign we would like to highlight is the Elixiaform redesign.
To be honest, I am not sure when this landed.
I just kinda saw it a week ago.
And I personally like it.
Gus has some a different opinion on it.
But generally, happened is that previously when you went to Elixir forum, you would have a list of all
the latest threats that were posted.
And whenever a threat was updated, when there was a new apply, it would just jump to the top of the queue
again.
And, you know, you might have some different filtering, but the default view was basically a long list
of threats.
And, yeah, they were not really sorted in any way other than when they were last updated.
And the new layout of the elixirform, when you just go to elixirform dot com, it shows you kind of trending
topics instead.
So they, you know, go one step higher where they say which big topics were discussed the most, you know,
got the most likes or views.
And what I like is the they also have different trending categories.
So they can say trending this week, this month, this year, three years, over three years.
I mean, you know, might not be that useful, but especially week and months.
That is cool if you just wanna get, like, an update on what has happened in the last week or in the last
months in the Elixir forum.
And then you see these very nice cards, which also give you information about, is this like an announcement?
Is this a blog post? Is this a talk a talk announcement? Or like an like a library update, that kind
of stuff.
So, yeah, I think you can also filter them on the left side by news, by, you know, blocks, and and and
so on.
So, you know, guess you have a different opinion, but what I personally like about this is that it gives
me a much easier access to the big topics that have been happening in the last week or months.
But you you said earlier you didn't like it that much, so so why is that? Well, I think that the UI looks
really great.
I think that Asen and the Elixir Forum team has done a really great job at at some of the UI updates
making it a nice fresh modern look.
I think the feature that you liked was that it doesn't show you the latest posts was actually one of
the ones that I didn't like.
Mhmm.
For the Nerfs meetup, when I make a new post in my existing thread about, oh, upcoming meetup next month
is, uh, whenever, it doesn't jump to the top.
It doesn't get that visibility.
Right.
And so that's a little bit of a downside in my opinion.
Yeah.
So it's harder to get the the eyeballs on on I actually I use that trick as well a little bit whenever
you had whenever you wanted to have a jump back to the top of the queue, you would just say, hey, I released
version whatever whatever, you know, and then Exactly.
This is which is kinda interesting.
But then it's mostly about getting the eyeballs back on your topic.
Yeah.
It is.
And I guess it's a it's a change of prioritization to the new things, the the or the the threads that
are actively engaged with a lot with the community right now.
And so and I think one of the use cases that I remember, um, reading on the justification for why they've
made some of these changes was that if you have casual users who are just driving by, they click on an
Elixir forum link, it's their first time there, they want to be able to see more relevant information.
Right.
I do actually know, I think it's forum dot elixirforum dot com.
Exactly.
It gets you back to the old view and then you can see everything like it was.
Yep.
There's a big button in the menu bar where it says view forum threads and if you click that, you go back
to the old view.
Yeah.
Right.
And so I know that this is one step or one phase in the in the progress for converting ElixirForm to
like a fully Elixir based platform which I think is a several years long initiative that it's still gonna
be a little while but I'm excited to see those updates and I do think that working towards that is a
is a great goal for the Elixir Forum team.
I agree.
So congrats to them for the redesign because it went off pretty it launched pretty smoothly.
Yeah.
I gotta say, I I like it.
I find it super useful, and thank you for the work on that.
Yeah.
The next item is that the Stack Overflow twenty six developer survey is now open on actually, it's already
open for a month.
So, you know, this is our first episode.
We need to catch up a little.
But about a month ago, the Stack Overflow twenty twenty six developer survey opens and that yeah.
People ask whether Stack Overflow is relevant anymore.
I I I think not.
I know with AI.
But the developer survey is always a nice thing to see which languages have been used more.
But to be honest, I'm not sure whether this like, now with Stack Overflow being less and less relevant,
people also receive these emails less frequently, you know.
So I'm not sure whether the latest news of as representatives I mean, they were never a hundred percent
representative of the industry, but at least you could see trends in them.
And, like so anyway, this year is it's still open.
The survey is still open.
So if you haven't already, go to the just Google or ask your your favorite AI for the link to Stack Overflow
survey twenty twenty six, and then you can Note, it's for human developers only.
Exactly.
So you can't have your AI fill it out, but Yeah.
You know, you can have it give you answers and then type them in.
About how much we all love Elixir.
Exactly.
Yeah.
I I I mean, I've been following them extensively, and every year, Elixir got, like, you know, zero point
three percentage point extras.
Like, that was nice.
But what it also showed is that Phoenix is, like, every single year is the most loved web framework.
You know, that's something that I always like to to post on the socials as a kind of brag.
Humble brag.
Yeah.
Next item, I'm gonna do this one.
It's from my good friend, Felipe Cabathu.
He announced x tori, e x dash tori.
And, actually, I know that Felipe, he has been working on that a year or two ago already, where, basically,
if you use Tori, which is a framework to build desktop app desktop apps, and they have Tori as like,
they have a Rust back end, which is written It's written in Rust, but it's the Tory Rust back end, and
that connects to your device, to your hardware, you know, and and that kind of stuff.
But then on top of that, you always have a front end framework, and usually that's React, the view, something
like that, or Svelte.
And Felipe, he went out and he added Live View to that as well.
So now you can write your front end with Live View, and then the back end is still Rust.
And this way, you know, you don't have to relearn, like, a rep framework, like Vue just to build desktop
apps.
You can just as well use Live View.
So I I thought this was really cool.
And it seems that now it's a proper library that you can you can use some tests.
So, yeah, maybe try this out for your next desktop app.
Now guys, this is something that is close to your heart, so the next item.
Really elixir related, not even I think it's just a cool thing.
They're definitely not paying us to to to put this here, but Augmentl, uh, launched a product called
Mouthpad, which I think is a really interesting device to look at and you might not know of them.
They're a pretty small company, pretty small startup, but it's kinda cool on pushing the barriers of
I think what human computer interaction can be.
So the mouth pad is a think of it like a retainer, like a like for your mouth.
So you put this retainer and it has a whole bunch of, like, sensors in it and you stick it in your mouth
and it turns your tongue into a cursor or like input device.
Mhmm.
Their use case that I've seen has been for anywhere like you need hands free control.
It can be for like surgeons mid surgery who obviously can't touch things because they need to keep sterile
environment or maybe like a pianist who's controlling their The sheet music.
Their app to show the sheet music or something like that.
The demos that they show on their website are very impressive in terms of the granularity of control
that you can have on it.
But I also wanna highlight this because I think it's a great exhibition of technology for persons who
might be disabled because those who can't use their their hands for various reasons now have a new way
of using an input device.
Right.
And so I think it's just a cool piece of technology for for the good of humanity.
I think this is a this is a very neat project.
So the link is gonna be linked down below.
Yeah.
But give them a check them out for some maybe some inspiration on what the future of technology might
look like.
Yeah.
I I saw this and and I was immediately very impressed by it.
I I also thought about people who have, you know, issues using their hands that who can, you know, only
use their mouth or things.
So this this allows them really to just move a browse like a a mouse.
Right? So it's kinda like a cursor they can control with their mouse.
It would be interesting.
And I think they have keyboard shortcuts and a whole bunch of Right.
You can customize it quite a lot.
They have a pretty extensive, like, SDK, I guess, for developing apps with this thing.
Oh, wow.
Okay.
That that is really just it doesn't run Nerfs, though.
Right? No.
Don't think it runs Nerfs.
That would be cool.
That would be cool.
But then I would need to ship a whole Raspberry Pi with it, so then you need to have a very big mouth.
It might work for, like, peoples and elephants, but not for humans.
Here we go.
Alright.
So now next up, I would like to introduce a little subsection of the news, which for me is something
that's also close to my heart, which is blog posts.
You might know my name from the blog posts that I've written over the last couple of years, and that's
how most people know it, me.
So I I thought it was it would be interesting to highlight just a couple of blogs and focus on blogs
in general in this section.
And a couple of blog posts I read in the last couple of weeks, The the first one was from Bart.
He is the creator of the Hologram library.
Hologram is, yeah, allowing you to write client side apps in Elixir.
Oh, sorry.
Is it Erlang? I think it's Erlang.
Oh, sorry.
But, anyway, it's it's basically if you wanna, yeah, build a client side app that runs in the browser,
but you, yeah, don't wanna use JavaScript.
Now you can use Hologram, which is written in either Erlang or Alexei.
And but wrote a a nice blog post about what he has been up to in the last year since the ElixirConfTalk
twenty twenty five.
He got a a sponsorship, you know, by company backers, so he has a good funding to build out Hologram
into the next into the next phase.
And, yeah, he just he just goes through everything he has built.
And it's not a super big blog post, but it gives you a good overview of what Hologram is, what the history
of it is.
So, you know, I I really appreciate that one.
Then He's been busy.
It's been cool to see some of the updates that he's pushed out because there's some cool stuff there.
True.
I saw him at elixirconf twenty six, right, where he also gave some updates again.
And every single year, it's just it gets cooler and cooler.
And I I hope that in the future, we can, like, have just client side Elixir apps running, and then they
connect back to your Elixir backend some way, you know, through channels or whatnot.
But then because people say, oh, Live View, you know, if what if the the connection breaks? Or what if
you wanna use it on your phone? Well, no, we have an answer.
Or we will have an answer.
Have you have you played with Hologram yet? I saw the the demo at the ElixirConf, so that was very impressive
as well.
But I haven't tried it out myself yet.
Okay.
We'll have to give it a go and and report back to the listeners because We will.
It's a cool project.
Exactly.
The next blog post is of an ex colleague and a good friend of mine, Ryan Zidago.
He wrote about actually, a couple of blog posts he wrote, but one of them that I liked a lot, which was
each test should have its own setup.
And in this blog post, he talks about yours Elixir test setup.
What's it? What do you call it? Like, routine or or to you know, how you set up these tests.
Basically, what he argues against is to have one huge setup block at the top of your test file where
you create every single record and and database structure you might need in the following tests.
Although most of the tests only use a subset of all these records.
Right? And and he argues it's better if you always create the the the setup for the test inside the test
and not in, like, a a higher level function like a setup block because this gives you maximum control
over what you need in this particular test.
If you scale it to thousands of tests across your code base, it will actually improve your runtime of
the test.
So that's also important in the long run.
And also it gives the reviewer and yourself, if you review it in the future, a better understanding of
which records and what shape they have in in this particular test.
So you you understand what is the setup, what is the actual test doing, and then what is it asserting,
and that gives you like a compact understanding of what it does.
Right.
I think I saw that his argument also was that it reduces the kind of the magic.
You you everything is all in one place so you can kinda see where values that you're asserting on might
come from and things like that.
So I think there's some good arguments there.
One one last yeah.
So this blog post is great but I also and and just this week or in the last week, he has posted six or
seven blog posts and they're all interesting.
So, you know, if you read this one, go back into his index and just have a look at the other ones as
well.
Good job, Ryan.
Great.
Keep it up.
And the next blog post that we wanna highlight is Mike Zornick, guarding against AI Drift.
I thought this was a really good one as well.
Um, Mike was talking about his setup and kinda tools for working with LLMs in a very structured way and
I think that this is personally, my opinion is that working with LLMs is best when you give them checks
and give them guardrails and give them clear, um, success criteria.
And Mike's blog post was about how you set up credo, how you set up, uh, various checks for code quality,
for, uh, correctness, like formatting, all these sorts of things so that your code, which is maybe generated
by an LLM, hopefully reviewed by you, is then of good quality and it doesn't kinda drift towards the
average of what AI can output.
Especially if you're starting with something like a handcrafted code base that you is near and dear to
your heart.
How do you make sure that the AI just doesn't slowly degrade that code quality over time? Exactly.
I thought this was a good read.
Yeah.
I like the in in the first block, he he makes this point very clear, which is he says, my worry isn't
any single generated change.
It's the long term drift.
And that's exactly the point.
Like, if you keep on building on your LLM code base with more LLM code, over time, you know, you will
have horrible test setups or, you know, horrible, like, long functions, not, like, repeating functions,
that kind of stuff.
And I I think in the years to come, we will see much more investment into these kinda static analysis,
like these codes so code based analysis tools that tells the, you know, hey.
The type system already is great for True.
For that if you've upgraded to one dot twenty, but I think you're right that there's a lot of investment
to be made here.
Exactly.
And and one item or one library that I have on my agenda to add is actually called axe slop, e x slop,
and that is also mentioned in this blog post.
And x slop is, yeah, giving you a couple of different tests that you can run, like, step what is it called?
Statistical no.
Well, tests code analysis that, you know, like, that gives you cyclomatic complexity, that kind of stuff.
And it it just, you know, says, yeah, this code is kind of too complex or too simple, and it, you know,
just gives you gives you LLM a test it can run and then improve on horrible code Right.
Just, know, to prevent slop, x slop.
Let's see.
Oh, yeah.
The last one is also from a person that I that admire and respect a lot, Johanna Larsen.
She wrote a couple of really great blog posts.
Again, I think these people, they they have a lot of time in the summer because all of a sudden, you
see that they've, you know, published one blog post a day.
They're productive.
They are Yeah.
So, Johanna, she wrote a blog about Elixir Cluster one zero one.
And, yeah, this is very much introduction to how Erlang notes are distributed, how they connect to each
other, how they communicate, and how you can send messages between them.
So this is a rather short blog post, but it gives you an interesting yeah.
Just, you know, like a highlight if you haven't looked into this before.
But she also has written quite a a few other blog posts in the last couple of weeks.
So one of them is how to stop Claude, sorry, saying load bearing.
That was interesting.
That was the funny one.
That was the funny one.
One of the better ones they were all great, but the one that I like particularly well was let libraries
be libraries.
It's also that, you know, you don't need to over engineer your libraries, you know, and also not everything
has to be a library.
Like, do these two things.
And one other one that I think she worked for she wrote for her her work, which is incident dot I o,
it seems.
She wrote a really great blog post about don't add a read replica until you've read this.
And that, yeah, goes through a a problem that incident had with the read replica and that basically,
yeah, prevented them from I think they had an outage because of this, but, yeah, don't don't quote me
on this.
But it's it's definitely an interesting read about yeah.
If you wanna add a read replica to, for example, reduce your query times or the workload on your right
database, you know, like, this is an like, an auto almost automatic decision.
Like, hey, let's add a read replica, then everything is gonna be golden.
Well, you know, just like anything in software engineering, not everything like, everything has trade
offs.
So this blog post really goes well into the the what are the trade offs of read read replica.
And, again, also, Johanna, she wrote a lot of different blog posts in the last couple of weeks.
So just go to the to her website or to the incident dot I o blog, and you can read them all.
Alright.
Perfect.
Well, we're gonna talk a little bit about upcoming meetups and conferences just to get these on your
radar.
And if you haven't already signed up for some of these, you might be interested and should grab some
tickets or register.
First up is the Nerves Meetup EU.
You're actually this is kind of the mini Nerves Meetup EU right now because on August twelfth, I will
be one of the hosts there and Peter is gonna be our guest speaker.
Yeah.
It's on August twelfth at seven PM CSD.
That's Central European Time.
So Amsterdam time.
You gotta check the time zones because it's a remote meetup.
We do it via Zoom.
Check the time zones.
You can sign up at nerfs meetup dot e u.
You'll find the links to upcoming meetups.
And, Peter, what are gonna talk about there? I'm gonna talk about don't get pawned by Mythos.
It's it's a little like, the title's outdated, but I update the content.
And it's about my security my vulnerability security research that I'm doing with the EF where we use
LLMs to find security vulnerabilities in in different libraries.
And so it kinda gives you an overview of where we stand right now in the industry, you know, what's the
latest and greatest.
Well, mostly not greatest.
What's the latest and worst? And what you can do to protect yourself against it.
Right? Like, basically, how will this whole AI apocalypse impact you, you know, in your daily work and
how you can protect this how you can protect yourself against it.
It's not necessarily focused on nurse, but I I will tailor it to what's nurse.
But surprisingly, nurse has been not surprisingly, but, you know, kudos to you because nurse hasn't really
shown a lot of vulnerabilities in my scans.
Actually, I haven't reported a single one so far, I believe.
You know? That's good.
Maybe until then I've met some But, I mean, it's still all all this security stuff still impacts Nerv's
users because we're still using libraries of the entire ecosystem.
One of the more interesting projects I think is the Nerves kiosk which turns a Raspberry Pi with a touch
display into a Phoenix Live View app that just shows on that display at startup.
So if there's Phoenix vulnerabilities then that's a nerves project vulnerability as well.
Exactly.
So, yeah, it'll be a really good time.
We usually have, you know, the presentations around thirty minutes and then we have some time for discussion
so that people can talk about what they're working on or get help on their nerves questions or just chat
with people to kinda meet others in the community.
We always have a really good time.
So, again, join us on August twelfth at seven PM c e s t, nerves meetup dot e u.
Awesome.
And then the next conference coming up that is also very dear to our hearts, and we already mentioned
it before, is Goldmeier in Warburg, Sweden.
It's run by our good friend Lars Wiegmann and his wife and team.
And, yeah, I'm also gonna be a speaker yet again.
I'm really on a speaking tour this year, but, yeah.
My talk there will be about LoRa, l o r a, long range radio frequencies, and how do you communicate through
them by text.
Actually, that's gonna be pretty cool.
But I'm excited for this talk.
It's gonna be a good one.
Me too.
I'm curious what I will come up with because I still need to get started on it.
How's the preparation going? Yeah.
Yeah.
Conference different development.
Right? It's great.
You know, I'm I'm very prepared to start.
But, yeah, the dates of Goldmeier is the September twenty eighth, so it's just a month which is the Monday.
And then it lasts until October third, which is the Saturday.
So it's a full week of conference.
Like, there's gonna be nurse conference and then the official Goldmire conference, but they have really,
like, one.
Then you have workshops before on the Saturday, which we wanna point out after the conference ends on
Friday.
On the Saturday, you have one extra day if you're interested in Ash.
So the Ash folks are gonna have the Ash conference.
And I'm also gonna manage a hackathon on the Saturday.
So if you just wanna stay a day longer in beautiful Warburgsburg, Sweden, and hack on any idea you have,
or, you know, just come along and hack on somebody else's idea, you're more than welcome to join us.
Just bring a laptop, you know, we're gonna have a great time.
There's gonna be maybe a little price in the end, but, you know, that's not what we're there.
We're just there for the good time and and just hacking together.
Yeah.
That's gonna be Goatmire.
It's gonna be a whole week of elixir festivities, essentially.
It's gonna be a really good time.
It's gonna be elixir nerfs week.
And and most importantly, this is the last Goldmire for at least next year.
There's not gonna be a Goldmire next year.
That already has been decided.
There might be a Goldmire the year after, so two years from now in twenty eight, but that has not been
decided yet either.
So if you wanna experience GoatMire, you have to come this year.
Right? The you you're taking a chance that it will happen again in two years, but, like, this year is
gonna be the last GoatMire for a while.
So you should come.
And let me tell you, if, like, last year, if it's if it's even half as good as last year, which I think
it certainly will be, think things have just gotten a little bit more I don't know.
He's added tick he's added seats.
Lars has added seats to the to the conference.
There's a few more spots available and the speaker lineup is looking really good.
There's workshop.
Like this whole thing is gearing up to be a really great conference.
Yeah.
So like there's gonna be FOMO if you don't go.
Hundred percent.
You better sign up.
You better sign up.
Yeah.
Then, guests, the next conference.
Yeah.
I guess we didn't do these in chronological order because before Gothmeyer, in September tenth eleventh
is ElixirConf US.
That is ElixirConf, so one of the two big ElixirConf events that happens every year, ElixirConf EU in
the spring, ElixirConf US in the fall.
There's workshops, there's trainings.
I will be there helping out with a little bit of the nerves workshop that's happening on the day before.
Mhmm.
So if you see me there, say hello.
But, yeah, it'll be a good time.
Go check it out.
I think it's elixirconf dot com Mhmm.
For tickets and registration.
Exactly.
September tenth and eleventh.
So where is it this year? In It's in Chicago.
Chicago.
Oh, yeah.
Chicago.
So you can eat a deep dish.
No.
What's it? Deep dish? It's a pizza.
Right? Deep dish.
Yeah.
Yeah.
Good.
I actually did one of those.
Is it really pizza? Is it? It's a cake.
It's a pizza cake.
It's more of like a casserole that just happens to be made out of pizza dough.
There you go.
Well, this is not a culinary polka, so let's just.
Man, we're gonna get some engagement from that statement.
I think some Chicagoans and New Yorkers are gonna Yeah.
Gonna I mean, if I could if I could have a pizza and just eat add depth to it like they did, I I would
be for it.
But I think this would last you for the entirety of the conference, so just eat one and you're done.
Yeah.
One of is probably okay.
Good.
And then the last conference we have is the CodeBeam EU.
It's gonna be in Harlem, which is close to Amsterdam.
And so you can stay in the area, just take a train to Harlem.
It's October twenty first and twenty second.
So, you know, this is probably gonna be the last big conference of the year.
And I'm yet again gonna speak together with my good friend, Johan Talmention.
We're gonna talk about the work on the vulnerability research we've been doing with the EF, and everything
about around, you know, vulnerabilities, AI security research, that kind of stuff.
So, yeah.
I think then I have something good one.
It should be another good one.
I always loved the Code Beam in Berlin.
This year, they're for the first time in Harlem and the Netherlands.
So Yeah.
It's just, you know, I can base I can basically see it if I look out the window.
I can see Harlem.
So I would have come anyways even if I wouldn't speak.
Yeah.
Very good.
Well, before that kinda wraps up what we were talking about news, blog posts, all that sort of stuff.
We're gonna move on to some more discussion topics on things that have been happening in the industry.
But before we do that, Peter, you are pretty famous for sticking a dad joke at the beginning of your
your blog posts.
So Right.
Please hit us with the dad joke before we get to the discussion.
Alright.
Well, just one.
You know? Just one.
Let's see.
I do like the one I have a speed bump for beer, but I'm slowly getting over it, which is nice.
Alright.
Oh, they're always corny.
Yeah.
It's like, is it a full laugh you're giving? Probably not.
It's a shuckle.
So I go for the shuckle.
Yeah.
The chuckle.
The chuckle.
That's that's a good metric.
Alright.
Yeah.
So now is the discussion time, you know.
So if you're interested in the news, you can shut off now.
If you wanna hear us ramble on about big topics that have been happening in the industry.
Yeah.
Keep on listening.
Right? Gus Okay.
The first topic, do you wanna introduce it? Shall I? Sure.
So Codeburg, which is the alternative git hosting site.
They're based off of for hey.
Forgeo? Forgeo? Yes.
Like forge j o.
Forge Joe.
Forjo.
Okay.
I don't really know how they they pronounce that, but No.
Forjo.
We're gonna go with that for now.
But CodeBurg is a hosted instance of Forjo.
Right? And they I think they're based out of Germany Yep.
Berlin maybe.
Yep.
So you can sign up with an account and push your git repos to there.
They have issues and pull requests and all that sort of same goodness that that GitHub has.
But the news with them is that they are now blocking LLM repos.
Yep.
And I think that the motivation behind this is that CodeBurg is a community run kind of nonprofit.
They run off of all their all their funding is just through donations.
Memberships.
Yeah.
And there's memberships too? Mhmm.
Okay.
But they have claimed that LLM repos are drain on their resources.
And I guess I mean, there's a lot of ethics and differing opinions on the AI and how it how it impacts
our community, but they've taken a hard dance against it.
So it's really it's kind of an interesting situation, and there's been certainly some backlash.
There's one interesting blog post.
I don't know who wrote that one, but Yeah.
It's kinda anonymous.
But so what CodePerk so what CodePerk did is they had a a membership vote.
Like, they had a a general voting what do you call it? Like, a general assembly of the members.
And they have been proposal one proposal in particular that just said we should block LLM generated repos
from being hosted on CodeBurk.
In the and then they changed the terms of used base of it, but they couple of problems with this.
So first of all, the problem with the term of used change is they are not a hard they don't take a hard
stance against LLM generated code.
So the for example, the Zig language, they moved to code codeberg a couple of months ago, and they have
adopted a much harder stance on LMs.
They basically say, you you should not use LLMs, and you can't even do research with LLMs to then write
code and contribute to the the Zig language.
So it's it's really kinda like if you know, all the way to, like, you can't use LLMs at all.
And if you do, then don't contribute to the Zig the Zig language.
Now, with the code per attempt term of use change, it is that they don't take a hard stance, but they
just say that mostly LM generated repos.
So not, like, a hundred percent or whatever.
Like, it it's kinda like a vague measure metric where they say mostly LM generated repos are not welcome.
And that's that's hard to measure.
Right? Unless Yeah.
I I what I understand is that if they're looking for that co authored by Yeah.
Plot line in your git commits Yeah.
Most likely.
But I don't know how they would check otherwise.
I mean, like a slop, you know, like a tool that looks at how many comments the LLM added because that's,
you know, like I mean, yeah.
And the existence of claud dot m d or agents dot m d files.
True.
Things like that.
But I think it's it's just an interesting issue at hand in how how we perceive LLMs.
I mean, one of the recent things that I've that that I saw as well was that Linus Torvalds, the creator
of Linux, the Linux kernel, he has stated that LLMs are a clearly useful tool.
And so, I mean, I guess the Linux kernel stands is that the code you commit, the code you push, the thing
the patches you wanna integrate into the Linux Mhmm.
They are they reflect the code.
They they they reflect you regardless of if an LLM wrote them or not.
Exactly.
And so LMs can be used as tools in Linux kernel development.
He's not banning that at all, but just use it responsibly.
And I think that's a relatively moderate But CodeBurg is is putting their foot down, having a hard stance
on no LLM generated repos.
Yeah.
And But the weird thing I I wanted to add a little because, like Yeah.
The the weird thing about this change with CodeBerg is not that they just state hard facts in terms of,
hey, you know, peep like, people who use LLMs you know, one argument could be people who use LLMs, they
generate more code than the average developer, and they push all of that to CodeBerg, which is a strain
on our resources.
Right? That that would be a perfectly valid argument where they say, you use physical resources by generating
code.
We don't want that.
We are, membership finance than an NGO in Germany.
You know, we don't have unlimited resources.
So please use a different hosting provider.
That I would understand.
But Right.
They made a they wrote a blog post after the the change went viral a little bit.
And the wrote a blog post with their kinda with the motivation behind this change.
And these arguments they bring here are not that clear cut, and and and they rather go into this, like,
do more anti AI level where, like, in the blog post that we are selling in the show notes, First, they
say, strange servers due to nonsensical crawling, which is that's the other problem.
Right? Like, they say they're crawling LLMs.
They go out and they read they come to our website.
They download their source code and they scan it.
Right? Like, these scanning crawlers.
That is I understand.
That is a a a physical and a financial strain on your servers.
It's a real problem.
It is a real problem.
But it's also not a problem with committed code Exactly.
Authored by LLMs.
Yeah.
So it's like, okay.
I understand this problem, but it has nothing to do with what people use your platform for.
So, I mean, then then it's it's already okay.
It's already clear.
It's it's not against, like, generated code.
It's against LLMs and AI in general.
You know? They don't like the crawlers.
They don't like hosted codes.
They're just anti AI in that case.
And then then it gets even weirder.
And and that kind of section of the blog post got the most feedback, I would say, on in the open source
community.
Because what they say here is that if you build your library or your open source code with LLMs, you're,
like, a development team of none.
They say you are just a single person kinda, like, writing or, like, having code being written for you
and then, like, throwing it out there, but you don't care about building a community.
You also don't have people interacting with your code.
Like, they also make a stance against single use software where they say you only build it for yourself
and you don't care to actually make it like a library that people use and build a community around it.
You know? And and they say that people who use LMS, they're like that.
Like, they don't they don't have a community.
They're not interested in building a community, and they only write single use software.
And they don't wanna have that in their community of code work, which if it was true, I would understand.
Right? You you kinda have a thriving community and you don't wanna have, like, people just, like, coming
in and not caring about anybody.
But I don't think that's true though.
Right? And it's not mutually exclusive either.
It's not that if your code is LLM generated, your community is dead.
There can be communities built around LLM generated code Yeah.
Just like handwritten code.
There's the yeah.
I don't know.
I think it's the problems that they experience and also I think that the the moral ethics around AI and,
you know, the IP usage, these are valid concerns, but the the way that they have justified this change
is a little confusing.
Yeah.
And and I read some commentary on especially this part where other people said, look, I've been writing
open source code before GitHub existed, and I have never had an an external contributor contributing
to my libraries.
You know? And and I think that happens a lot where you just write a library, you put it out there.
Right.
People might download it and use it, but you will never hear from them, and they will never contribute.
Like, know, they might create issues, like an issue on GitHub or but they will never send you a pull
request with contribution.
Right? That that was these people's arguments.
Like yeah.
And out of and and CodeBurk themselves, actually, they also kinda they split from Gitea Gitea.
It's a different Yeah.
Yeah.
Another open source.
They also like yeah.
They were also like, hey, let's just split the community and and use our own thing, know, instead of
building our own community around something new.
Like, they also kinda reuse that.
So Well, all this does come at an interesting time with GitHub outages.
People are not happy.
I mean, GitHub is probably a perfect example of AI generated, LLM generated code does have a impact on
resource usage and performance of the entire platform.
They have had a huge increase in the number of daily users or I don't know if all this if you count If
you the bots as users.
Right? The CTO posted this screenshot, but then people kinda added numbers to it because, of course,
they always take away the labels Right.
And the the access.
And people and then you saw over three years, they had an increase of, like, four or five x of pull requests
and commits over three years, which is like Right.
That is something that, especially the Microsoft infrastructure, should hold up against.
Right? But then they use Azure, so I think that's just an Azure issue.
Please tell us more about your Azure We're beautiful.
Oh, well, you should probably go save it and look at your the recording of your previous ElixirConf talk
Yeah.
Because I think you did explain your feelings towards Azure on The the mapping the cloud, I I you know,
like, that's a talk I think that will come out soon.
But, yeah, I go into detail about my experience with Azure and and others.
Yeah.
Looking at the time, should we, like, move on to the next Maybe maybe I'll introduce this because it's
it's, like, related to my vulnerability work.
Again, like, we speak about AI a lot right until now.
Right? But this is something that I just want to quickly highlight, which is there is an a whole ongoing
situation between Hugging Face and OpenAI, where Hugging Face claims an OpenAI then they I say, they
committed to it.
No.
They agreed to it.
They said it's true.
That Hugging Face said that an OpenAI model hacked the infrastructure, basically.
And they found this hack before even OpenAI was aware of it.
And then they, uh, Hugging Face, they, uh, announced it.
And then OpenAI looked at it and was like, hang on.
That's probably us.
So and then OpenAI said they own investigation, and they found out, yep, they, um, they were benchmarking
a new model.
And the new model decided instead of doing the actual benchmark, it would be easier to hack this other
system that is out there, Hugging Face, because it it thought that Hugging Face would have the answers
to the benchmarks that it was running.
And yeah.
Okay.
You can you can do that.
And the the problem here was that it should have been very heavily sandboxed, the the model, from using
the Internet.
It was not allowed It was, wasn't it? It was.
Like, it was sandboxed, and it only had access to the Internet for a couple of things through a proxy
software.
So, everything was channeled through this one proxy software.
And the proxy software was configured to only allow, you know, a couple of, like, package managers and
GitHub, that kind of stuff.
Right.
But first, the model hacked that proxy software with a vulnerability that was not disclosed before.
And then it gained access to the entirety of the Internet, and then it decided to go into Hugging Space
infrastructure, like hack it, and spin up thousands of jobs on it, you know, to try to find answers to
run the benchmark, that kind of stuff.
Was kinda like, hey.
There's a lot of compute power over there.
How about I just use it? Right? So that was kinda scary when I read that, to be honest.
You know, just for the capacity, for the capability of that model, but then also about the alignment
where even if the model, you know, saw that this would have been a possibility, it should kinda have
stopped at one point.
No? Yeah.
That it didn't I mean, it unleashed itself.
It's kinda like we're getting into the the sci fi dystopian AI takes over the world kind of things here.
But, yeah, it's It's scary.
No? It's it's like these news that that scroll over the wind the TV wherein you have an apocalypse movie
and you introduce a scene you introduce the, you know, the the world, like and then you see this TV,
yeah, AI just hacked everything kinda.
The scary is, like, one of the kind of spooky things, but we also what's it called? Like, where where
we put, like, ourselves or human beings into machines, and we expect them to also be anthromorphizing
or so is the word.
No.
Yeah.
It is.
Yeah.
It's a word.
Right? Anthropomorphizing.
Anthropomorphizing? It's a hard word to say.
Anthropomorphizing.
Anthropomorphizing.
That one.
Good word.
That's the one.
So that word is, you know, something where we when we look at what a model does, we might be like, oh,
it's sentient.
You know, it has consciousness.
But so the Is it though is it just the token generator? Yeah.
But then, know, I saw a funny Blue Sky post where somebody is like, you know, as the paper clip maximizer
killed me, somebody said, yeah, but does it actually understand that it's killing you? Doesn't There's
the line for sentience.
Right? Yeah.
Like, eventually, if it kills you, it kills you.
It doesn't matter whether sentient or not.
You know? It's yeah.
But but one last item here, then we move on, which is one scary thing is that this model of OpenAI, it
also left notes on OpenAI servers or Hugging Face servers, one of them, for the next model with instructions
about how to free itself from the the sandbox.
I didn't hear that one.
Yep.
That's pretty funny.
That is pretty funny.
Yeah.
Funny in a scary way.
Hello's humor.
Right? Yeah.
Yeah.
Yeah.
One one item, because people were skeptical about the news, because everything about scariness is marketing,
obviously, for OpenAI and and others.
They they were like, yeah.
This is just marketing between Hugging Face and OpenAI.
I I don't believe so, to be honest, because Hugging Face, they released first.
I mean, okay.
They could have organized it, but the reaction that I saw from both companies, and now that they also
work together with the government in the US to kinda understand what happened.
I I think this is more serious than just a let's make a funny marketing thing.
And I don't Sam Altman would fall into the same trap that Dario from Anthropic fell in, which is to over
hype the scariness of a model like Mythos, and then have the US government just, you know, crush down
on you and say, well, if it's so scary, you shouldn't release it to everyone in the entire world.
Right? And and Right.
And then Trophy was like, what? I think they learned their lesson by that example from overhyping it.
Exactly.
But but, you know, an incident like this might have that effect on on the perception of of these models
coming out.
I I do believe so, especially with missiles now, you know, having laid the groundwork of the government
being afraid of that and having export controls in place.
Like Right.
I think this this will just keep on coming.
Yeah.
And and Sam And the crazy part.
Will know how to circumvent these relations better.
But, yeah.
It's something we have to take into cons oh, yeah.
Sorry.
One last item before we yeah.
Because The crazy part.
The crazy part tell tell it.
Yeah.
What's the crazy part? The crazy part is that that that when who was it? It was Hugging Face Mhmm.
Their their team was trying to contain the the breach, I guess Yeah.
And and the the AI going rampant.
They tried employing also, I think, Claude and Codex Yeah.
Models.
Not Codex For Fable.
Yeah.
Fable.
Mhmm.
It was only fable? I think so.
Okay.
But they were employing these models from Anthropic to contain the the breach and the damage and diagnose
what was happening and how it was how it was doing what it was doing.
And when they tried to to to do this diagnosis, they kept getting downgraded to Opus or Yeah.
To lower tier models or or just shut off like this is not something that Fable supports because you're
in one of those restricted Yeah.
Topics.
Um, so they had to switch to was it Kimi? No.
The Chinese five point two.
It's an open weight Oh, GLM five point two.
Yeah.
I don't think Kimi was out at that point.
Well, it was, but it was Kimi two point six.
It was not that useful.
Like, GLM five point two was the most useful one at that point.
And, yeah, they they spun up their own instance of that model on their own servers.
And then they, yeah, used it to analyze what was happening, the logs and and the vulnerabilities and
everything.
Right.
Which is And this is Hugging Face.
They have almost unlimited resources to do this.
But imagine if it happened to a smaller company.
True.
Like and then you get get get rate limited or not not even rate limited, just downgraded performance
by Claude when you're trying to diagnose what happened to your infrastructure.
Yeah.
I mean, it shows you two things.
Right? One of them is that the the smaller people well, Hugging Face is not small.
But whoever is not a Frontier Lab, they will always kind of fight with one hand tied behind their back
because they they don't get access to these latest and greatest and unrestricted models that, you know
Right.
Potentially an attacker has.
In this situation, it wasn't OpenAI attacking Hugging Face by accident, but still.
So Accident.
Yeah.
Accident.
So so they have that issue.
And I I see that also with my works, like, yeah, I wanna use Fable, but I can't use it as useless for
vulnerability research.
Even if you're, like, a verified researcher on on the platform, like, it always downgrades you to four
point eight.
Codecs, like the GBT five point six, it's just what I use nowadays because it doesn't have these restrictions
and it's it's just really good.
Yeah.
So that's the the first thing.
You will kinda always be at a disadvantage against the the big boys.
And and the second thing is also as Europeans, because Hugging Face is from France.
Right? We're also kinda we would have both hands almost tied behind our back because, you know, even
if we wanted to, like, Hugging Face could not get access to Mythos, or at least I believe they they most
likely haven't, because they're a European company, And and they have Yeah.
European employees and everything.
Like, if they were fully a hundred percent American, they they could have gotten access to Mythos.
I I know that from my work because we have exactly this issue that some communities, open source communities,
they do have access to Mythos because they're in America and they have American employees.
And then Right.
We in in Europe, we can't get access because just because we're European.
So then And that's the second problem.
Other interesting thing on the the difference in in jurisdiction that I think is gonna come out of this
is when this happens again in the future because I think that's a when not an if.
Yeah.
Who's responsible? The model is taking actions on its own.
It's not like they gave it a prompt to you know, hack it out sandbox and go after Hugging Face.
Yeah.
But this is these are all open questions that that we need to spend some time as humans thinking about
Yeah.
And figuring out the answers to.
But it'll be interesting.
That's true.
I mean yeah.
So so enough about this topic.
Let's just you know, because we're already in one hour, let's dive into the very last one, which I found
interesting, which is basically if you think about us as software engineer and how our work will look
like in five years with AI being better and better, like, yeah, how will our job even look like? And
how should we react to what is happening in the industry? You know? Should we be anti AI and just try
block it as much as possible, or should we embrace it, or should we use it as a tool, like anything in
the middle? So if you look at the math the math community, mathematicians, they are they're even more
impacted by this, because their work inherently was just brain matter, you know, finding the solution
to a really complex problem.
And Right.
We've been seeing so many change or, like, disproves of conjectures of, like, math theorems, you know,
of of, like, established math foundations.
All of a sudden, they were just disproven by GPT five point six, you know, when, like, some young kid
started off that process while it was playing League of Legends.
That actually happened.
And then, you know, it it finished a couple rounds, and I looked back at the output, and the the LLM
was like, yeah.
I found a count example to this really well known theorem, you know, from, like, eighty years ago.
So this is, like, the the biggest breakthrough, the craziest breakthrough as you can see in an industry,
I would say.
And and now the question is so there's a really nice block by the Spectrum Magazine, and they wrote a
really nice blog article about, yeah, what it means to be mathematician when AI does the math.
Right? And what you kinda see in that community could teach us how our community will look like or how,
you know, could embrace this.
And they're basically three different angles.
Well, probably four different angles in in the math community as this blog post states it.
One of them is obviously the anti AI thing where you say, don't wanna use AI.
I don't wanna use LLMs.
I'm gonna do everything like I have to with pen pen and paper, you know, not maybe it was like computer
help assisted, but not LLMs.
Right? So this will be the anti.
But interestingly, in the blog post, that opinion has almost zero surface.
Like, it has almost zero mentioning.
Because if you're a mathematician and you don't wanna use LLM, you're out of a job pretty quickly.
You know? So that's like case one.
But then the the three groups that this blog post points out is, first of all, AI is a tool, you know,
and that's also what Linus Tewold said, where you are are the human who has the understanding of the
problem, who can kinda design a solution, who can, like, point at the right direction to potentially
solve a solution, and then the AI becomes your assistant at executing your your vision, so to say.
And this is a rather popular belief in the community, especially by, like, the I also didn't know this
name before, but Terence Tau.
That's his name.
He is apparently a boy genius who is the yeah.
Terence Tau from University of California.
He's probably the smartest person on the planet, at least in the mathematician community.
And if if mathematicians say that this guy is the smartest guy, that tells you a lot about how smart
he is.
Right? Yeah.
So so Terence Tau, he is very much in the AI as a tool group and says, yeah, we just need to learn how
to use it, and we will still be the designers of the solution, and then the AI does the the grunt work
of finding and proving stuff.
The second group is kinda AI as a partner group where the there's it's not a clear cut human being the
orchestrator or the, you know, yeah, the the supervisor and then AI being the assistant.
Is rather a we are collaborating with the AI.
We're kind of on same levels.
You know, if the AI wants to try different angles to your problem, we just let it do it and, you know,
it might also bring problems to us that we then counter verify.
And and the most important thing here, and that is a funny thing because also in the blog post, they
point out the motivation of mathematicians quite well, which is not, like, to get the the biggest finding
or to have, you you know, the, like, the biggest name in the industry.
It's very much about solving problems and just knowing what the freaking answer is to a problem.
I find it so interesting that they're like, I I don't care whether I I will publish this.
I just wanna know the answer.
You know? I I imagine a lot of our community resonates with that as well.
Exactly.
That's how I think a lot of us might have gotten into computers is Yeah.
How does this magic glowing box work? And and that is interesting part because I think that is different
for different people.
Like, if you're Right.
If you're a person who whose problem you wanna solve is to write really nice and clean code or solve,
an algorithm issue, you know, like sorry.
These are two things.
If you wanna write clean code, yeah, maybe LLM, like, they will take that away from you, so you might
feel threatened more because you just don't get to write code anymore.
But if you are very much into computer science and you wanna figure out the most optimal, sorry, algorithm
and efficient algorithm to solve this big data issue or, like, you know, speed issue, I think you you
will very much still be, well, allowed or able to do that in the future.
You know? Right.
You will have an LLM.
Novel discovery is just assisted Exactly.
In it's it's a partner, a collaborator in your Yeah.
Your your Your toolbox.
Yeah.
It's like Yeah.
You know, it's like I I come up with an idea for an algorithm that might fix this issue, that might solve
this issue, And then I I either write it I write it myself and have it verified by the LLM, or I have
the LLM write a first version of it, and then I can tweak it until it's efficient.
Yeah.
Right.
And things like writing tests and benchmarks, these can be Yeah.
Helped and automated by Exactly.
By AI.
Yeah.
But at the end of the day, I think there still is a lot of the novel idea generation for this type of
work coming from the human brain.
Exactly.
I I I believe so as well.
And, actually, I have, well, proof, quote, unquote, by Jose Valin, you know, who who says that, yeah,
sure, he can use an LLM to write applications.
So, yeah, you look at Chris McCord who generates, like, a web app using LLM on stage while giving a target
at ExitComp twenty five, um, you know, which was impressive back then already.
But, know, it's a web app.
Like, it's just a web app, quote unquote.
And for Jose, it's writing the language and and writing, you know, like the type system.
That that is something that no one LM has seen before, especially for Alex here.
Yeah.
It's it's very much, how you call that, like, groundbreaking work where nobody else has seen it before.
Right.
And he can't use LM so that he says he tried and it's just garbage.
So he he like the Elixir language, he very much writes himself still because it's just the most efficient
way.
Yeah.
I thought it was funny at ElixirConf EU twenty twenty six.
Jose and Chris had a little race Yeah.
To see who could implement a fix for a bug, I think it was ECTO or something like that quicker.
And I think Jose won.
He won.
Yeah.
The manual implementation.
So He was faster.
And more correct.
Yeah.
Yeah.
It's just a more it's it's a sign that we shouldn't be outsourcing our brainpower to the LLMs because
we still have a lot of value there.
Exactly.
And and as I was saying, like, the blog had these three two three groups, AI as a tool, AI as a partner,
and then the last one is just AI as an oracle.
And I think that is kind of the the future idea of AI where software engineers, particularly, they're
afraid of, and they don't wanna move into this direction, which I agree with, which is that, you know,
AI is just the autonomous researcher.
Like, the autonomous developer, product manager, whatever.
You just let it do its thing because it's so much smarter than us.
And what what we as humans only do is, you know, we take the the answers from the Oracle and, you know,
we spread them in the world kinda like the Jesus and followers kind of situation.
So so that is, you know, also what I don't think will happen, and I don't think I I would like this to
happen.
Yeah.
Yeah.
It's interesting though with this article how they break it down and think of it very structured in in
in the use of of AI in these areas.
Because I do think that I mean, one thing that we humans are very good at is putting classifications
to things.
Mhmm.
Things in our in to make sense of the world.
So this might be one framework that you could use to see how AI as an industry is going to impact work,
the nature of work itself.
Exactly.
Yeah.
And and I I think also if you have personal fears about how it will look like in the future, maybe look
at these groups and see where it falls into which group does it fall.
Right? Like, are you afraid that you won't have a say anymore? Then you're, like, AI is an or is an Oracle
kind of group, or, like, then you fear that we'll fall into that group and, you know, like, that it might
be valid, might not be valid, but it might also help you to focus on the other groups then to see where
you stand to to understand, well, I still I love doing this orchestration work, you know, as AI is a
tool.
Like, I use I like to use it as tool.
That's very much my camp, where I'm like, I wanna build this application.
And I wanna kinda brainstorm the architecture and the design, you know, and how things are being done.
But then I don't wanna write the actual code.
Right? Right.
So this is You you understand it at a very fundamental level, but not every line.
You don't you don't you don't hand type every character in Exactly.
Your code file.
I mean, at the end of the day, if you have a large enough code base, you will forget how parts of it
work anyway.
Right? Need to go back into the code, understand it.
But you will you will always keep in mind the architecture that you've chosen and the software design
that you've chosen, because that's high level enough for you to be, you know, simple simple enough to
understand.
Yeah.
Alright, guys.
Well, that is a lot of AI.
I don't think that we want to always be talking AI AI AI.
Yeah.
But it is I mean, of course, it's hard to avoid right now.
It's just a big thing in our community and having impacts in all sectors of life, I would say.
So it's Might be unavoidable.
Unavoidable.
Might be unavoidable.
But we'll definitely bring more topics and exciting things to talk about in future episodes.
Exactly.
Well, thank you everyone who has stuck around until the end.
We appreciate it a lot.
Yeah.
This has been the first episode of Macro Mayhem.
This has been fun.
It's been fun.
Next episode will be in two weeks.
Yeah.
Keep us follow us on wherever you get your podcasts.
You can also follow us on well, me or Gus or both of us on Blue Sky or LinkedIn.
Or we have an account for the We do have an account.
The the podcast.
So if you're on Blue Sky, it's macro mayhem dot f m.
Exactly.
So Our website is also macro mayhem dot f m.
There will be the latest spot episodes.
I'm I will There's not much there.
Right? It it will once we release this one, it will be Hopefully, by the by the time that we release,
we'll have some content.
Exactly.
And and I will also release this to YouTube so that if you wanna watch us on on YouTube there is actually
video to this.
Of beautiful faces.
Of two beautiful faces.
Yeah.
I even got my Elixir hat on.
So Oh, nice.
I should have been wearing my Nerf shirt.
There you go.
Next time.
Which brings us to the end of the show.
Thank you very much for sticking around.
And, god, I can't I can't do the outro of of thinking, Alexey.
No.
We have to do something else.
That's it, folks.
This has been a wild mayhem hour, so I don't know.
We'll have to think of Yeah.
But that's it folks.
Good night and see you next time.
Cheers.
Cheers.
